A 6-DOF articulated arm with 5 kg payload, 710 mm reach, and ±0.02 mm position repeatability fits a typical bench-top machine-tending or light-picking cell, as listed in a mid-2026 Articulated Robot Solution spec sheet [S6].
Selection in practice reduces to five hard numbers — payload (kg), reach (mm), pose repeatability (mm), cycle time (s), and DOF count — plus three soft constraints: mounting (floor/ceiling/wall), controller openness, and integrator ecosystem [S1][S6].
Where articulated arms sit in the industrial robot taxonomy
Articulated arms use all-revolute joints in a serial kinematic chain — base yaw, shoulder pitch, elbow pitch, and a 2- or 3-axis wrist — giving 4 to 7 DOF, distinguishing them from Cartesian gantries (all-prismatic), SCARAs (R-P-P plus one wrist revolute) and collaborative robot variants that share the same topology but add force-limited joints and rounded geometry. [S2]
The general 6R topology is the de facto cell-robot standard because the all-revolute chain packs the largest dexterous workspace into the smallest footprint, and a late-2026 reference design still specifies a six-revolute serial chain with three coaxial wrist shafts (outer, intermediate, inner) for tooling rotation [S4].
Decision criteria: five numbers that gate 80% of the shortlist
Payload is checked first, then multiplied by a 1.5–2× gripper-and-cable margin so a cell rated for 5 kg nominal load rarely lifts more than 2.5–3.3 kg of part-plus-end-effector in continuous duty [S6].
Reach is the radial distance from the J1 axis to the tool flange centre (Point P); a 710 mm envelope suits a single-machine-tending cell, while 1400–2000 mm arms target automotive spot-welding lines and large-format palletising.
Pose repeatability is quoted at the tool flange under ISO-conditional test loads, and the 2026 mid-payload class sits at ±0.02 mm while compact 4-axis arms typically quote ±0.01–0.015 mm and heavy-payload 6-axis arms relax to ±0.05 mm [S6].
Cycle time is the standardised 25-300-25 mm pick-and-place beat; 0.5 s is the current benchmark for 5 kg / 700 mm class arms, and drops to 0.3–0.4 s only on low-payload SCARA-class machines [S6].
DOF count is the gate for dexterity: 4 axes cover top-loading bin-picking; 6 axes are the default for welding, painting, and machine tending; 7 axes add a redundancy that helps reach behind obstacles in body-in-white cells.
Variant comparison: compact, mid-payload, and heavy-payload classes
Compact articulated arms (≈5–8 kg payload, 700–900 mm reach, ±0.02 mm repeatability, ≈40 kg body mass) are the right answer for electronics assembly, lab automation, and CNC machine-tending where a linear guide track is not justifiable [S6].
Mid-payload arms (10–20 kg payload, 1400–1700 mm reach, ±0.03–0.05 mm repeatability, ≈150–250 kg mass) target arc welding, sealing, light deburring, and machine tending of mid-size CNC lathes; the wider J2/J3 envelopes (+85° / -125° and +185° / -55° on a 2026 reference unit) keep the wrist clear of fixturing [S6].
Heavy-payload arms (50–800 kg payload, 2500–3100 mm reach, ±0.05 mm repeatability) are the right answer only when the part and gripper together exceed 30 kg; below that threshold the cost, footprint, and energy bill do not amortise against a mid-payload unit.
Mounting options further split the shortlist: floor-mount is the cheapest, ceiling-mount is preferred in paint cells where drip fallout must fall clear of the arm, and wall-mount only works on the slim-wrist variants where the J2 envelope has been derated for gravity loading [S6].
Dynamics, dynamics, dynamics: why link length and inertia decide cycle time
Link lengths and joint types are normally chosen by intuition and workspace plots, but the dynamic criterion — simplicity of the Generalised Inertia Matrix (GIM) — is what actually gates cycle time, path accuracy, and servo stability, per a peer-reviewed dynamic-model selection study [S2].
The same study shows that a GIM with few off-diagonal terms (or one rendered diagonal by kinematic redesign) lowers floating-point cost in the inverse-dynamics loop and reduces CPU time per control tick, which directly translates into faster and more stable point-to-point motion [S2].
For buyers, the practical read-through is to demand the inertia tensor at the wrist (Jx, Jy, Jz) and the rated payload-inertia ratio (typically 10:1 to 20:1) on the datasheet, then size the gripper so its reflected inertia stays under one-tenth of the arm's wrist rating.
Who should NOT pick the mainstream 6-axis articulated arm
Do not pick a 6-axis articulated arm when the trajectory is planar, top-down, and high-speed — a SCARA delivers 0.3–0.4 s cycle times at lower cost, and a Cartesian gantry is cheaper still above 2 m of travel. [S3]
Do not pick a 6-axis articulated arm when the cell is shared with humans and the part is light; a force-limited collaborative robot trades raw speed and payload for power-and-force limiting and rounded geometry, and is the only variant that drops the light-curtain budget.
Do not pick a 6-axis articulated arm when the floor footprint must stay under 0.25 m²; even a compact 5 kg / 40 kg arm needs a 0.5 m × 0.5 m baseplate, and a crossed-roller-guide Cartesian stage on a column gives the same reach in roughly one-third the floor area.
Controller openness and integration: the part buyers forget
A 2026 reference controller ships with an Intel Core i5-3610ME, 2 × 2 GB DDR3, 500 GB HDD, and an EtherCAT master port — enough headroom to host a vision pipeline alongside the motion loop, but the CPU is now two generations behind current industrial PCs and the DDR3 footprint constrains vision-buffer size [S6].
Buyers should treat the controller as part of the spec: ask for the fieldbus (EtherCAT is the de facto standard in 2026), the high-level language API (ROS 2, Python, or vendor-proprietary), and the digital I/O count, because retrofitting these later is harder than swapping a robot.
For lines where multiple brands coexist, the articulated robot spec sheet is also where the encoder protocol, the safety I/O wiring (STO/SS1 on dual-channel inputs), and the EtherCAT slave stack version are pinned; mismatches on any one of these three create integration cost that exceeds the arm price differential.
Use cases by variant, with concrete numeric envelopes
Machine tending of a small CNC lathe: 5–8 kg payload, 700–900 mm reach, ±0.02 mm repeatability, 0.5 s cycle, floor-mount, mid-open controller with EtherCAT and 16 DI / 16 DO [S6].
Arc welding of steel sub-frames: 10–20 kg payload, 1400–1700 mm reach, ±0.03 mm repeatability, hollow wrist for the dress pack, and a 7th-axis positioner integrated through the same EtherCAT master — the class of work where a sizing method like the one in electric actuator sizing and selection: a working engineer's guide crosses over to the rotary drives that index the part.
Palletising of 25 kg cartons at 10 cycles/min: heavy-payload arm (50+ kg), 2500+ mm reach, ±0.05 mm repeatability, with vacuum or clamp end-effector; a 6-axis arm beats a gantry here only when the pallet pattern changes frequently, otherwise a linear gantry on a linear guide beam is the lower-cost answer.
Selection shortlist logic, and what to verify on the datasheet
Step 1 — write the payload with gripper and cable margin (×1.5–2.0) and the worst-case reach from J1 to the farthest pick point; the smallest arm that still satisfies both is the baseline. [S3]
Step 2 — verify pose repeatability at the actual payload (vendors quote at rated load; under-loaded arms do not improve), then confirm cycle time on a path that resembles the real trajectory, not the 25-300-25 mm benchmark.
Step 3 — confirm J2 / J3 envelopes match the fixturing; a 2026 reference unit quotes J2 = +85° / -125° and J3 = +185° / -55°, and any cell where the wrist would have to fold past -125° at J2 is mechanically impossible on that arm [S6].
Step 4 — pin the controller (EtherCAT, ROS 2, safety I/O, vision headroom) and the mounting (floor / ceiling / wall) on the PO, because the integrator's commissioning hours are the line item most often blown by a soft spec.
Step 5 — request a 2-day cell simulation from the vendor with the actual 3D CAD of the fixture; arms that look right on paper frequently collide at the wrist between J3 and J4, and this is the cheapest place to catch it.
Trackable signals over the next two quarters: the migration of mid-payload controllers from Intel Core i5-3xxx-class CPUs to current-generation industrial PCs, and the broader rollout of ROS 2 native drivers in place of vendor-proprietary APIs; both shift the total cost of integration more than any incremental reach or repeatability gain.